EP2850402B1 - Estimation of drift in a solar radiation sensor - Google Patents

Description

Field of the invention

The present invention generally relates to sensors for evaluating solar radiation, such as pyranometers or reference cells, and, more particularly, to the estimation of the drift and the calibration of a sensor of this type.

Disclosure of prior art

Sensors for evaluating solar radiation are increasingly used with the development of solar installations, whether for the production of heat or electricity. These sensors are used, in particular, to measure the radiation (watt/m ² ) on the premises of the solar energy production installations. These measurements are used, among other things, to estimate the solar resource or monitor the correct operation and energy production of solar power plants.

There are essentially two categories of radiation sensors known: sensors consisting of reference photovoltaic cells, and pyranometers which are specific sensors dedicated to measuring the radiation from the sky (solar radiation). A pyranometer is a heat flux sensor that measures the radiation received.

All these radiation measurement devices provide information (for example, a current, a voltage or a digital word) representative of the radiation they receive.

A recurrent problem with the use of solar radiation sensors is that the measurements drift over time. This drift requires regular recalibration or calibration.

Today, to be recalibrated, the sensors are generally dismantled and then sent back to the factory or to workshops to be subjected to reference radiation and to correct the coefficients of an algorithm applied to the measurements and providing the real radiation from a raw measure.

The ISO 9847 standard describes devices for calibrating solar collectors when returning to the workshop or by comparison with other collectors specially brought to the site by a maintenance service. These devices are not suitable for calibration or drift correction, on site, continuous over time and automatic.

The documents US 7,166,825 , US 7,576,346 and US 2009/0012731 describe systems for recalibrating radiation sensors mounted in space vehicles. The systems described make it possible to calibrate the sensor by comparing the measurement made on the object concerned (a given astral body) with a measurement made on a reference object for which the radiation is known with precision (the sun for example whose level of radiation is known provided one is outside the Earth's atmosphere).

Such systems are not applicable to sensors on earth, since solar radiation can no longer be used as a reference due to the variability of the transmission of radiation by the atmosphere depending on the meteorological conditions.

The complexity of recalibrating a radiation sensor means that many instruments are not re-calibrated quite frequently (in practice a few times a year at most), which affects the accuracy of the measurements and the estimation of the production capacities of solar power plants. In addition, the proliferation and location of solar power plants (for example in private homes) makes such recalibrations more complex. Finally, cost considerations may mean that this operation will be neglected in certain cases.

The document EP-A-2211300 describes a method for predicting the electrical production of a photovoltaic device and proposes the diagnosis of a photovoltaic installation by comparing an actual production with an estimated production. It is a measurement of the actual electrical production of the photovoltaic modules.

The article " Monitoring and remote failure detection of grid-connected PV systems based on satellite observations", by A. Drews et al., Science Direct, Solar Energy 81 (2007), p. 548-564 , describes a fault detection system in photovoltaic panels, based on satellite observations, and which aims to avoid the use of reference cells or a pyranometer.

Summary

Thus, an embodiment of the present invention aims to propose a technique for estimating the drift of a radiation sensor which overcomes all or part of the drawbacks of the known techniques.

Another object of an embodiment of the present invention is to provide a calibration technique suitable for on-site operation.

Another object of an embodiment of the present invention is to propose a technique for estimating the drift and calibrating without external intervention.

Another object of an embodiment of the present invention is to make it possible to increase the frequency of calibrations of a radiation sensor.

Another object of an embodiment of the present invention is to provide a solution requiring no structural modification of the radiation sensor.

The invention is defined as claimed in claim 1 and the claims dependent thereon.

Brief description of the drawings

• la figure 1 illustre, de façon très schématique, un exemple de système d'estimation de la capacité de production d'énergie solaire à l'échelle d'un territoire ;
• les figures 2A et 2B sont des exemples arbitraires d'allures de rayonnement solaire reçu par un capteur de rayonnement ;
• la figure 3 est un schéma bloc illustrant des étapes d'un mode de réalisation du procédé d'étalonnage d'un capteur de rayonnement ; et
• les figures 4A et 4B illustrent le fonctionnement du mode de réalisation de la figure 3.

These objects, characteristics and advantages, as well as others will be explained in detail in the following description of particular embodiments made on a non-limiting basis in relation to the attached figures, among which:

• the figure 1 very schematically illustrates an example of a system for estimating solar energy production capacity at the scale of a territory;
• them figures 2A and 2B are arbitrary examples of patterns of solar radiation received by a radiation sensor;
• the picture 3 is a block diagram illustrating steps of an embodiment of the method for calibrating a radiation sensor; and
• them figures 4A and 4B illustrate the operation of the embodiment of the picture 3 .

detailed description

The same elements have been designated by the same references in the different figures. For the sake of clarity, only the steps and elements useful for understanding the embodiments described have been detailed. In particular, the exploitation of the measurements of the radiation sensor to estimate the energy production capacities of a plant has not been detailed, the embodiments described being compatible with the usual exploitation of measurements provided by a sensor of solar radiation. Moreover, the practical realization of a radiation sensor and the conversion of the measurement signal have not been detailed either, the embodiments described being, here again, compatible with the usual radiation sensors equipped with calculation means .

Various examples of implementation and realization are presented below. Regardless of the name given to these examples (embodiments, example, variant, etc.) and the qualifiers used (for example, preference, etc.), only the parts of the description included within the scope of the claims are part of the present invention, the other examples being useful only to highlight specific aspects of the invention with respect to what is not part of it.

The figure 1 illustrates an example of a centrally operated solar energy installation for power distribution management purposes.

Multiple solar power plants 12, each consisting of one or more sets of photovoltaic panels (solar panels) 122, and of an energy management and conversion system 124 (including in particular an inverter) are distributed over a territory F The system 124 is generally associated with a radiation sensor 2 (for example of the pyranometer type). This sensor measures radiation in the Earth's atmosphere.

The energy supplied by each power station 12 can, for example, be injected into the electricity distribution network of the territory F and, at the same time, the various power stations 12 send data (links 126) to a computer system 3 centralizing the management of the energy. This system 3 generally comprises a network control and operation room, equipped with one or more computers 32, one or more screens 34 and one or more databases 36 receiving information from the various power stations.

The system 3 enables one or more operators to manage the distribution of energy from information, known a priori or collected from the various power stations, which they view(s) on the screens. This information includes, among other things, the geographical location of the power plants (map 341), information 342 on production and demand and, for example, information on the radiation 343 received by one or more several plants, allowing a comparison with expected production.

The representation of the figure 1 is an illustrative example of a possible application of the embodiments which will be described. This illustration is of course simplified, the described embodiments not relating to the exploitation of the measurements obtained by the solar radiation sensors to predict an energy production capacity or to manage the distribution. The exploitation of the production capacities calls upon usual techniques, generally independent of the nature of the source of energy.

A particularity of the production of solar energy is however the need to predict, in particular on the basis of meteorological conditions, the production capacity of solar power plants in relation to other sources of energy. Another peculiarity is the dissemination of power plants with low production capacity (less than 10 kW) throughout the territory.

These particularities mean that the reliability of the radiation measurements carried out at the level of the power stations themselves is particularly important.

In addition, the dissemination and the large number of solar power plants make it more difficult to return pyranometers or other sensors to the workshop for recalibration.

One could think of using portable calibration devices and organizing control visits to the various plants on a periodic basis. However, this considerably increases the cost of calibration and therefore the cost of energy production. Moreover, it is not possible to visit the small-capacity power plants frequently, which delays the consideration of any deviations accordingly.

As part of the monitoring of a group of solar energy production plants, it may be necessary to forecast production for the near future. In this case, most systems rely on field measurements to feed artificial intelligence procedures. The Solar radiation measurements are of course taken into account. Consequently, the presence of aberrant measurements due to the bad calibration of the sensors considerably handicaps the performance of the forecasts. There is therefore a real need for a system for estimating and correcting the drift of the sensors, all the more so if the system operates in a homogeneous manner on all the power stations. Of course, other applications are conceivable, such as, for example, monitoring the performance ratio of photovoltaic power stations, weather forecasting, etc.

où α et β sont des coefficients de la fonction affine déterminée pour l'étalonnage du capteur.The calibration of a radiation sensor, initially or in operation, consists in determining the coefficients of an affine function applied to the measured values. This function corrects the measurements and provides an radiation value. By denoting G _MES the measurement taken and G _COR the corrected measurement, at a given time t, the corrected measurement is obtained by applying a relationship of the type: $${\mathrm{G}}_{\mathrm{HORN}}\left(\mathrm{you}\right)=\mathrm{\alpha }.\left({\mathrm{G}}_{\mathrm{MY}}\left(\mathrm{you}\right)-\mathrm{\beta }\right),$$

where α and β are coefficients of the affine function determined for the calibration of the sensor.

In practice, the drift of the coefficient β is negligible and we are concerned here only with the drift over time of the coefficient a.

It has already been proposed, for example in the article " An evaluation method of PV systems" by T. Oozeki, T. Izawa, K. Otani and K. Kurokawa (Solar Energy Materials & Solar Cells 75 (2003) 687-695 ), a method for evaluating the performance of several interconnected photovoltaic modules. This method evaluates the power produced daily by the plant over the course of a month. We then seek the minimum coefficient by which it would be necessary to multiply the so-called clear weather radiation, that is to say in the absence of cloud, to encompass the measurements made by the sensor. Such adaptation does not take place in real time and requires measurements over several days. This allows to define a kind of "effective peak power" of the plant, but this does not imply an estimate of the radiation received.

The figures 2A and 2B illustrate two examples of radiation patterns (Watt/m ² ) obtained over time (h) throughout a day by a solar collector. It is these radiation measurements that must be usable in order to then estimate the energy production capacities of a solar power plant. During a normal day, that is to say, without variations in radiation other than those linked to the path of the sun, therefore at the time of day, the curve approximately follows a bell curve. In practice, there are always disturbances (clouds, objects, dirt (bird droppings), etc.) acting as a temporary screen between the sensor and the sky. The figure 2A illustrates the case where minor disturbances appear. The figure 2B illustrates the case where, for a few hours, the sensor is partially shaded but enough to cause the radiation received to drop significantly.

The calibration method proposed below is based on an estimation of the drift of the sensor using periods during which the sensor is in a clear sky condition (absence of clouds).

A radiation model is then used providing a theoretical curve of the radiation, in clear sky conditions, for the zone in which the collector and the solar power plant are positioned. This theoretical curve can take into account other parameters such as the date, the inclination and the orientation of the sensor.

The value of the correction coefficient α (which also represents an estimate of the drift) to be applied to the measurements is adapted, for the current day, according to an exploitation of the measurements of one or, preferably, of several previous days.

The picture 3 is a block diagram illustrating steps of an embodiment of the method for calibrating a solar radiation sensor.

The figures 4A and 4B are timing diagrams illustrating the operation of this method.

The method is implemented by a digital processing circuit of the microprocessor type, programmed to implement the various steps which will be described. Digital processing circuits and memories are used which are usually fitted either to the radiation sensor 2 itself, or to a possible device (124, figure 1 ) management which receives the information from the radiation sensor 2, or more generally any computing device capable of communicating with the sensor.

A daily pattern 51 of radiation in clear weather of the zone in which the sensor is placed is stored (block 41, TEMPLATE) in the processing device. This storage is carried out, for example, during the installation of the solar power plant and the collector. As a variant, the model is calculated on the fly, which makes it easier to take into account parameters other than location (date, inclination, orientation, etc.).

Theoretical models allowing to estimate the radiation in clear weather on a given region according to the date, the time, the longitude, the latitude, the orientation of the panel or the sensor, the inclination, albedo, etc. are known. It is assumed that the model of block 41 takes into account all or part of these parameters and, preferably, all of them.

Such models are, for example, described in the article " On the clear sky model of ESRA - European Solar Radiation Atlas - with respect to the Heliosat method" by C. Rigollier, O. Bauer and L. Wald, published in the journal Solar Energy Vol. 68, No. 1, pages 33- 48, in 2000 .

These models provide, depending on the date of the year, a bell curve 51 ( figure 4A ) indicating the radiation in Watt/m ² according to the hour (h) of the day. At night, the radiation is almost zero.

To estimate the drift and determine the coefficient α to be applied to the measurements so that, in the event of clear weather, the radiation measured corresponds to the radiation of the model, one seeks to determine, during a day of real measurements carried out by the sensor , periods in which it was in a clear weather situation.

For this, the model being established on a daily cycle, the rate of the radiation measured by the sensor during a day is stored (rate 53, figure 4A ). In practice, to optimize the control of the calibration of the sensor, the measurements of the day which precedes the instant of the calibration are stored (block 42, DAY-1). We can consider that at midnight (24h), we have the complete daily cycle of the past day. The following processing applies to the data of this past day.

The real pace obtained 53 is then compared (block 43) with the clear weather model 51. The objective of this comparison is to determine one or more time windows or ranges 54 during which the real pace 53 can be considered as corresponding to exposure in clear weather.

In practice, the comparison is carried out on numerical values, insofar as the sensor generally provides discrete values over time (it provides a value for each measurement). The comparison is preferably carried out on all the measurements taken by the sensor, which improves the reliability of the result or, as a variant to save calculation resources or consume less, on only one measurement out of a predefined number of measurements (for example , one measure out of two, one measure out of four, etc.). It is also possible to provide several consecutive aggregated measurements (by their mean, their median, etc.). The choice of the number of measurements taken into account depends, among other things, on the measurement frequency of the sensor. If it takes a very large number of measurements (for example, every second or several per second), it is possible not to take into account all measurements without losing too much information. If, on the other hand, the frequency of the measurements is lower (for example, every minute), it is preferable to take into account all the measurements for the calibration.

For each measurement (sample), we seek to determine whether or not it corresponds to a clear weather instant (that is to say at a time when the sun is not masked). For this, one can, for example, estimate that the times when the curve of the measurements compared to the theoretical curve in clear weather is relatively smooth (with a variation threshold on the standard deviation of the variations over a sliding window) and that the measured values are quite high (and this, over a sufficient time interval around the considered moment).

The inventors have observed that the difference between the real radiation over a period of clear weather with respect to a model of this radiation directly provides an at least approximate value of the coefficient α of the correction to be made to the sensor for its calibration.

In fact, considering periods of clear weather, i.e. during which the sensor should give values corresponding to those of the model, the corrective formula (1) above should give the value of the model from the measured value. By neglecting the coefficient β which is, in practice, close to 0 and therefore all the more negligible as the radiation is high (case of clear weather), we see that the correction to be made comes down to the ratio of the value given by the model to the measured value.

Thus, the ratio of the measured instantaneous value G _MES to the instantaneous value G _TEMP given by the model is evaluated (block 44, d=G _MES /G _TEMP ). This positive quantity d is all the lower when the sky is overcast and is theoretically equal to unity if the sensor is perfectly calibrated and the sky is clear. The figure 4B illustrates the pattern 55 obtained over the time window 54. This pattern gives an indication of the defect of the coefficient α (in fact of its inverse).

As a variant, it is possible to evaluate the inverse ratio 1/d, that is to say the ratio G _TEMP /G _MES which directly gives the value of the coefficient a. Whichever ratio is used, that ratio represents an estimate of the sensor drift.

In order to avoid taking into account insignificant periods (for example at night when the radiation is in principle close to 0) or strong variations (dawn and dusk), the determination of the clear weather time window (block 45, WINDOW) takes place within a 56 day period. This period of the day is chosen, by way of example and arbitrarily, as corresponding to a period in which the radiation according to the clear weather model is greater than a given value, typically 50 Watt/m ² . It is in fact generally considered that the periods in which the radiation is between 0 and 50 Watt/m ² correspond to dawn and dusk. This absolute threshold of 50 Watt/m ² is of course adapted according to the location. As a variant, the selection of the day period can be made from a calendar and the indication of the time.

In the case where several time windows are taken into consideration, the average of the values obtained over the different periods is performed, for example.

The calibration is preferably carried out during the second part of the night, that is to say between midnight and dawn of the following day. The simplicity of the determination allows daily calibration. If no acceptable time window is available on a given day, the calibration is postponed to the next day.

The reliability of the sensor calibration correction depends, among other things, on the determination of the clear sky moments. The more efficient the detection of clear weather moments (that is to say, the more false positives and false negatives will be discarded), the better the determination of the coefficient a.

To detect periods of clear weather, it is possible to search for the time windows for which the ratio d (between the measurements and the theoretical model) shows relatively small variations and a sufficient level. For example, if the ratio d, over the time window around a given instant (for example, 1 hour before and 1 hour after the chosen instant), has a standard deviation lower than a threshold value and belongs to an interval deemed reasonable, the moment chosen may be considered as a moment when the sky is clear. The range of plus or minus one hour is an example that can be modified.

The inventors have observed that the choice of the width of the time window can be modified by taking into account (among other things) the sampling of the data considered. The shorter the time interval between two measurements, the smaller the window size can be. If no acceptable time window is available on a given day, the calibration is postponed to the next day.

The calculations on time windows allowing the detection of clear weather instants are carried out iteratively on each of the samples. Such iterative determination is particularly suited to software-driven digital processing systems.

According to a simplified embodiment, it is considered that the estimation of the coefficient α for the day considered, which represents the correction value to be applied to the affine function of the sensor, corresponds to the inverse of the median value of the coefficients d present in the time window(s) (block 46, α=1/d). The adaptation of the sensor (block 47, SENSOR ADAPT) is carried out from the coefficient α obtained.

In practice, the estimation of the coefficient α varies from one day to another, whether due to a drift in the operation of the sensor, this drift being progressive, due to a small quantity of time instants clear during the day, or following inaccuracies in the detection of periods of clear weather (for example periods of cloudy gray weather but stable throughout the day).

Thus, a weighting of several determinations of the coefficient α is preferably carried out as a function of one or more factors.

Preferably, the importance given to the various factors is itself weighted. By way of example, the inventors consider that the seniority of the day is the most important factor, the reliability of the day coming second and the absolute level of the day being the least important.

Other weightings may be considered to make the determination even more reliable. For example, we can take into account the deviation of the current day from the average of the previous days.

Taking the example of the age of the day, the reliability of the day and the absolute level of the day, we can, for example, use the following factors.

A factor takes into account a weighted average of the coefficients of the previous days by assigning a weight that is all the lower to the coefficient of a given day as this day is older than the current day.

Another factor takes into account the estimated reliability of the day considered. This reliability corresponds to the number of reports d taken into consideration on the day considered, which amounts to classifying the days, or the coefficients a, according to the number of periods of clear weather on which the value is based. The greater this number, the greater the weight given to the coefficient α considered.

Yet another factor is the level of radiation obtained. This amounts to emphasizing the days in which the values calculated for the ratio d are the highest.

Determining the weight to be assigned to the coefficients α of the different days, taking into account the above variants, amounts to calculating the product of the factors for each day. The weights are obtained by normalizing the results so that the sum of the weights is equal to 1. In practice, one can choose not to consider that a number of the days corresponding to the most recent.

In practice, the above steps are carried out by successively processing the samples of the day in question. This processing can be expressed as follows.

We note i, the time when the rank of the sample during the day, and j the day (j=0 for the current day, i.e. the day at the beginning of which the evaluation from the measurements performed is calculated, j=1 for the day on which the measurements are made, i.e. the day before the current day, j=2 for the previous day, etc.).

We denote by x, a time range (in number of samples) around the current sample defining an interval (time range) during which the measurement must be more or less stable for an absence of disturbance to be considered (cloudy Or other). A threshold y of the ratio d(i) is fixed below which the standard deviation of the ratio calculated over the time range must remain for the radiation to be considered to have remained "stable".

A threshold z for determining measurements that are too dark or aberrant is fixed. For example, if d(i) is less than 1-z, the sky is considered "dark" at time i, if d(i) is greater than 1+z, the sky is considered "abnormally clear" at time i (aberrant measurement). Following a long period of drift, a sensor may return values outside the interval [1-z, 1+z]. It is however possible to guard against an inappropriate elimination of these values by normalizing the values G _MES from the estimation of the coefficient α of the previous day. In this case, once the check has been made, the inverse operation will be carried out so as not to disturb the rest of the calculation.

• la plage horaire i-x à i+x est comprise dans le "jour", c'est-à-dire soit la mesure G_MES est supérieure à un seuil (par exemple 50 watt/m²) dans toute cette plage, soit la plage exclut la nuit, l'aube et le crépuscule, ou les deux ;
• l'écart type des rapports d(i) sur la plage i-x à i+x est inférieur au seuil y (par exemple, le seuil d'écart type est choisi entre 0,03 et 0,1, typiquement de l'ordre de 0,06) ; et
• la valeur du rapport d(i) est comprise entre 1-z et 1+z (par exemple, le seuil z est compris entre 0,2 et 0,6, de préférence de l'ordre de 0,4).

A ratio d(i) is taken into account for the calculation of the median (taken into account in the 55 rate) if:

• the time range ix to i+x is included in the "day", i.e. either the G _MES measurement is greater than a threshold (for example 50 watt/m ² ) throughout this range, either the range excludes night, dawn and dusk, or both;
• the standard deviation of the ratios d(i) over the range ix to i+x is lower than the threshold y (for example, the standard deviation threshold is chosen between 0.03 and 0.1, typically of the order of 0.06); and
• the value of the ratio d(i) is between 1-z and 1+z (for example, the threshold z is between 0.2 and 0.6, preferably of the order of 0.4).

où N représente le nombre de jours passés pris en compte (de préférence entre 10 et 30) et pj représente le poids donné au jour de rang j. Il s'agit donc d'une moyenne pondérée des coefficients α_j.The coefficient α ₀ is obtained according to the following relationship: $${\mathrm{<figure-callout\; id="0"\; label="\alpha "\; filenames="imgf0002.png"\; state="\{\{state\}\}">\alpha </figure-callout>}}_0=\frac{{\displaystyle \sum _{\mathrm{I}=1}^{\mathrm{NOT}}\left({\mathrm{p}}_{\mathrm{I}}\cdot {\mathrm{\alpha }}_{\mathrm{I}}\right)}}{{\displaystyle \sum _{\mathrm{I}=1}^{\mathrm{NOT}}{\mathrm{p}}_{\mathrm{I}}}},$$

where N represents the number of past days taken into account (preferably between 10 and 30) and pj represents the weight given to the day of rank j. It is therefore a weighted average of the coefficients α _j .

In a simplified embodiment, only the seniority of the day is taken into account. For example, we assign weights p _j corresponding to the rank of the day (p _j = j).

et ${\mathrm{p}}_{\mathrm{j}}^{\mathrm{e}}$

respectivement donnés à l'ancienneté du coefficient α_j, au nombre de données sur lesquelles s'appuie le coefficient α_j et à la valeur du coefficient α_j d'après la relation suivante : $${\mathrm{p}}_{\mathrm{j}}=\frac{{\mathrm{p}}_{\mathrm{j}}^{\mathrm{t}}\cdot {\mathrm{p}}_{\mathrm{j}}^{\mathrm{d}}\cdot {\mathrm{p}}_{\mathrm{j}}^{\mathrm{e}}}{{\displaystyle \sum _{\mathrm{j}=1}^{\mathrm{N}}\left({\mathrm{p}}_{\mathrm{j}}^{\mathrm{t}}\cdot {\mathrm{p}}_{\mathrm{j}}^{\mathrm{d}}\cdot {\mathrm{p}}_{\mathrm{j}}^{\mathrm{e}}\right)}},\mathrm{avec}:$$

$${\mathrm{p}}_{\mathrm{j}}^{\mathrm{t}}=\exp \left(-10\cdot \frac{{\mathrm{n}}_{\mathrm{j}}^{\mathrm{t}}}{\mathrm{N}}\right);$$

$${\mathrm{p}}_{\mathrm{j}}^{\mathrm{d}}=\exp \left(-7\cdot \frac{{\mathrm{n}}_{\mathrm{j}}^{\mathrm{d}}}{\mathrm{N}}\right);\mathrm{et}$$

$${\mathrm{p}}_{\mathrm{j}}^{\mathrm{e}}=\exp \left(-4\cdot \frac{{\mathrm{n}}_{\mathrm{j}}^{\mathrm{e}}}{\mathrm{N}}\right),$$

où :

• ${\mathrm{n}}_{\mathrm{j}}^{\mathrm{t}}$
  
  désigne le numéro de l'estimation α_j rangée de la plus récente à la plus ancienne (par exemple 1 pour α₁, 2 pour α₂, jusqu'à N pour α_N) ;
• ${\mathrm{n}}_{\mathrm{j}}^{\mathrm{d}}$
  
  désigne le numéro de l'estimation α_j rangée suivant le nombre de d(i) pris en compte pour son calcul (1 pour la valeur de α_j basée sur le plus grand nombre d'instants i considérés comme "temps clair" le jour j, N pour la valeur de α_j basée sur le moins grand nombre d'instants) ; et
• ${\mathrm{n}}_{\mathrm{j}}^{\mathrm{e}}$
  
  désigne le numéro de l'estimation α_j rangée suivant le niveau de la valeur α_j (1 pour la plus grande, N pour la plus faible).
• Les nombres 10, 7 et 4 des formules (4) à (6) sont des exemples de facteurs d'importance donnée aux poids ${\mathrm{p}}_{\mathrm{j}}^{\mathrm{t}},{\mathrm{p}}_{\mathrm{j}}^{\mathrm{d}}$
  
  et ${\mathrm{p}}_{\mathrm{j}}^{\mathrm{e}}$
  
  . Plus le nombre en facteur de l'exponentielle est élevé, plus le poids est important dans la moyenne de la formule (3).

In a preferred embodiment where the three factors mentioned above are taken into account, the weight p _j is obtained, by combining the weights ${\mathrm{p}}_{\mathrm{I}}^{\mathrm{you}},{\mathrm{p}}_{\mathrm{I}}^{\mathrm{}}$

and ${\mathrm{p}}_{\mathrm{I}}^{\mathrm{e}}$

respectively given to the age of the coefficient α _j , to the number of data on which the coefficient α _j is based and to the value of the coefficient α _j according to the following relationship: $${\mathrm{p}}_{\mathrm{I}}=\frac{{\mathrm{p}}_{\mathrm{I}}^{\mathrm{you}}\cdot {\mathrm{p}}_{\mathrm{I}}^{\mathrm{}}\cdot {\mathrm{p}}_{\mathrm{I}}^{\mathrm{e}}}{{\displaystyle \sum _{\mathrm{I}=1}^{\mathrm{NOT}}\left({\mathrm{p}}_{\mathrm{I}}^{\mathrm{you}}\cdot {\mathrm{p}}_{\mathrm{I}}^{\mathrm{}}\cdot {\mathrm{p}}_{\mathrm{I}}^{\mathrm{e}}\right)}},\mathrm{with}:$$

$${\mathrm{p}}_{\mathrm{I}}^{\mathrm{you}}=\exp \left(-10\cdot \frac{{\mathrm{not}}_{\mathrm{I}}^{\mathrm{you}}}{\mathrm{NOT}}\right);$$

$${\mathrm{p}}_{\mathrm{I}}^{\mathrm{}}=\exp \left(-7\cdot \frac{{\mathrm{not}}_{\mathrm{I}}^{\mathrm{}}}{\mathrm{NOT}}\right);\mathrm{and}$$

$${\mathrm{p}}_{\mathrm{I}}^{\mathrm{e}}=\exp \left(-4\cdot \frac{{\mathrm{not}}_{\mathrm{I}}^{\mathrm{e}}}{\mathrm{NOT}}\right),$$

where :

• ${\mathrm{not}}_{\mathrm{I}}^{\mathrm{you}}$
  
  designates the number of the estimate α _j row from the most recent to the oldest (for example 1 for α ₁ , 2 for α ₂ , up to N for α _N );
• ${\mathrm{not}}_{\mathrm{I}}^{\mathrm{}}$
  
  designates the number of the estimate α _j row according to the number of d(i) taken into account for its calculation (1 for the value of α _j based on the greatest number of instants i considered as "clear weather" during the day j, N for the value of α _j based on the smallest number of instants); and
• ${\mathrm{not}}_{\mathrm{I}}^{\mathrm{e}}$
  
  designates the number of the estimate α _j arranged according to the level of the value α _j (1 for the highest, N for the lowest).
• The numbers 10, 7 and 4 in formulas (4) to (6) are examples of factors of importance given to weights ${\mathrm{p}}_{\mathrm{I}}^{\mathrm{you}},{\mathrm{p}}_{\mathrm{I}}^{\mathrm{}}$
  
  and ${\mathrm{p}}_{\mathrm{I}}^{\mathrm{e}}$
  
  . The higher the factor number of the exponential, the greater the weight in the average of formula (3).

The sum of the weights p _j is normalized to 1 (the average being made by formula (3)). Therefore, formula (2) becomes: $${\mathrm{<figure-callout\; id="0"\; label="\alpha "\; filenames="imgf0002.png"\; state="\{\{state\}\}">\alpha </figure-callout>}}_0={\displaystyle \sum _{\mathrm{I}=1}^{\mathrm{NOT}}\left({\mathrm{p}}_{\mathrm{I}}\cdot {\mathrm{\alpha }}_{\mathrm{I}}\right)}.$$

An advantage of the embodiments which have been described is that it is henceforth particularly simple to calibrate a sensor in real time according to the drifts which it undergoes. This considerably increases the reliability of the solar radiation measurements.

Another advantage of the embodiments described is that their implementation does not require any structural modification. existing solar radiation sensors, the estimation of the drift, then the calibration are carried out from an interpretation of the measured and stored values. It will be noted that the estimation of the drift of the sensor, or the determination of the corrective value a, can even be carried out remotely from the sensor, in the device 124 for managing the solar panel, or even on a remote server.

Various embodiments have been described. Various variations and modifications will occur to those skilled in the art. In particular, everything that has been explained in relation to a direct measurement of the radiation can be carried out using magnitudes representative of this radiation. For example, it may be the current flowing in the measuring element of the sensor or any other quantity representative of the instantaneous radiation. Moreover, the practical implementation of the embodiments described is within the reach of those skilled in the art based on the functional indications given above using computer and programming tools.

Claims (10)

1. Method of estimating the drift of a solar radiation sensor (2), wherein said sensor is a pyranometer or a reference cell associated with photovoltaic panels, and wherein:
   the radiation measured (GMES) by this sensor in its conditions of use and a radiation model (51) are taken into account, using instantaneous ratios (d, α) of measured radiation values (GMES) to values (GTEMP) given by the model (51), or of values (GTEMP) given by the model (51) to measured radiation values (GMES);

   the model (51) delivers an estimate of the radiation expected in the case of a clear sky and takes into account the geographic location of the sensor, the latter being located in the terrestrial atmosphere;

   the measured radiation (GMES) is compared (43) with the expected radiation (GTEMP), delivered by the model (51), during periods corresponding to a clear sky; and

   the estimate of the sensor drift is calculated by taking into account prior estimates of the drift.
2. Method according to claim 1, wherein the estimate of the sensor drift is calculated daily.
3. Method according to claim 1 or 2, wherein an instantaneous ratio (d) is taken into account if the time is comprised within a time range (54) during which the ratio variation is smaller than a threshold.
4. Method according to any of claims 1 to 3, wherein a ratio (d) is taken into account if its value is comprised between two thresholds.
5. Method according to any of claims 1 to 4, wherein a ratio (d) is taken into account if it corresponds to a daytime period.
6. Method according to any of claims 1 to 5, wherein the estimate of the drift is obtained from a weighted average of estimates calculated during the previous days.
7. Method according to claim 6, wherein the weighting takes into account the distance in the past of the days taken into account.
8. Method according to claim 6 or 7, wherein the weighting takes into account a reliability coefficient assigned to the estimate of the considered day.
9. Method according to any of claims 6 to 8, wherein the weighting takes into account the value of the estimate.
10. Method of calibrating a solar radiation sensor, wherein a correction coefficient (α) to be applied to the measurements is obtained from an estimate of the sensor drift according to any of claims 1 to 9.

Images